"Prof. ( "Dean of Valuation"), this was such an insightful and engaging session and the "reevaluating the story part"! The way you broke down DeepSeek's potential impact on the AI landscape was eye-opening. I really liked how you framed it as 'Story Break, Change, or Shift?'-it made me think deeply about how we often overlook the nuances of disruption. Also, the idea of 'disruption for the disruptors' was spot on. It’s a reminder that no one is immune to change, no matter how innovative they are. Thanks for making these complex topics so relatable and for always tying them back to the fundamentals. Looking forward to more of your thought-provoking analyses!"
I have to add to the video that the H800 GPUs that are used to train DeepSeek are still Nvidia GPUs. The moat of Nvidia is not in the GPU chips, but in CUDA. IMO DeepSeek has not shown that you don't need supercomputers for the training. None of us normal people can afford to build a cluster that is required to train DeepSeek. An H800 costs $20k and they used 2048 of those. What they have actually shown is that there is absolutely NO MOAT around LLMs because any company with a big enough budget can simply distil the models from the output. The real issue aournd "AI" has always been copyright and nobody is talking about it. We have no copyright for the AI models themselves. At the same time, they're trained on copyrighted material and nobody seems to care because they expect AI to change the world. The moment they realize that it won't change the world, everyone will want to get their money for the copyright violations. Then, these models will become several factors more expensive. This is the real issue that needs to be discussed.
My old professors told me that when they were young, it was thought that computers were such a powerful but expensive machine, that large countries like USA and USSR would need a few of them each. To a lesser extent, it was thought that training AI were so expensive, only a few very well funded companies like Microsoft backed OpenAI could do it. If we get to a point where medium sized organizations could afford to do it, that's a huge game changer.
Deep seek post mortem explained so well and very useful. This is helpful for investors both retail and institutional. Thank you Professor for distilling out the sound wisdom and knowledge from the noise and distractions.
Have to say the separation in the run of the mill and the high end usage of AI was insightful as well as the "cute" ones. I am interested in the innovative use cases of AI that can make a small but positive change in people's lives and make them think more. Maybe a red-pill/blue-pill AI is the need of the hour.
Did you watch the video? Ineternet has allowed us to look up anything and decreased our reasoning skills. Try to work it out yourself with the information you already have
Deepseek ended AI monopoly / oligopoly power as it's opensource. I would add to your model assumptions the effect of a US govt clampdown on Nvidia's "Singapore" sales Great point about the blind spots. Having spent time in silicon valley, it's clear their version of AI abundance is somewhat ignorant of how the economy works Thanks for an excellent discussion Prof
Deepseek reverted closedAI back to OpenAI. Now anyone can wait for them to spend 100s millions of dollars to train frontier models so we can siphon the model with destillation
You forgot about the chip manufacturers and the manufacturers of the equipment to manufacture chips (your AI players story). Too much hate on physically capital intensive companies!
25:45 : The value of the “rest” (which includes gaming) is up about 300 B vs the September valuation , partly compensating the ~40% (or 550 b) drop in the AI business. Would anyone know why that is the case ?
It's rumored that huawei is developing a chip equal to nvidia's flagship h100 at a significantly lower cost. This could democratize the high end AI market as well.
They are still a few generations behind and if they don't have access to the processes of ASML because of export restrictions it may take a while. On the other hand with tariffs on EU companies by USA and increased isolationism, Europeans may be motivated to lower restrictions to China and have a more balanced approach.
Hi, Didn't Elon say Tesla will need to spend 500 billion dollars on compute for training in the latest earnings call? Then there's other companies who will enter the robotics and humanoid space, this not rule based, I think long term the Nvida story is still justified.
AI for big 7 is already making rewards, I know Big7 companies have been able to cut head count over 1000s because AI does what those head count were doing.
if this is a shift, then nvida will have a larger market. the current players will still suff their datacenters with chips because they still need to train the models. but additional players will show up because they use better models now on normal consumer cards. maybe this is also a tailwind to amd because there are also competitive chips aiming more on the consumer market
Deepseek required the OpenAi model to be built (see OpenAi lawsuit). The massive spending of OpenAI and others will be required for further development
Where did OpenAi get their data ? It is like a thief report to police that his car he stole was stolen by someone. Hypocrite.
"Prof. ( "Dean of Valuation"), this was such an insightful and engaging session and the "reevaluating the story part"! The way you broke down DeepSeek's potential impact on the AI landscape was eye-opening. I really liked how you framed it as 'Story Break, Change, or Shift?'-it made me think deeply about how we often overlook the nuances of disruption.
Also, the idea of 'disruption for the disruptors' was spot on. It’s a reminder that no one is immune to change, no matter how innovative they are. Thanks for making these complex topics so relatable and for always tying them back to the fundamentals. Looking forward to more of your thought-provoking analyses!"
I have to add to the video that the H800 GPUs that are used to train DeepSeek are still Nvidia GPUs. The moat of Nvidia is not in the GPU chips, but in CUDA.
IMO DeepSeek has not shown that you don't need supercomputers for the training. None of us normal people can afford to build a cluster that is required to train DeepSeek. An H800 costs $20k and they used 2048 of those.
What they have actually shown is that there is absolutely NO MOAT around LLMs because any company with a big enough budget can simply distil the models from the output.
The real issue aournd "AI" has always been copyright and nobody is talking about it. We have no copyright for the AI models themselves. At the same time, they're trained on copyrighted material and nobody seems to care because they expect AI to change the world. The moment they realize that it won't change the world, everyone will want to get their money for the copyright violations. Then, these models will become several factors more expensive.
This is the real issue that needs to be discussed.
My old professors told me that when they were young, it was thought that computers were such a powerful but expensive machine, that large countries like USA and USSR would need a few of them each.
To a lesser extent, it was thought that training AI were so expensive, only a few very well funded companies like Microsoft backed OpenAI could do it. If we get to a point where medium sized organizations could afford to do it, that's a huge game changer.
😊😊😊
Lol😅😅😅
One can rent training hours from data centers
Is it certain that throttled GPUs were used, and not high spec ones sourced secretly?
Deep seek post mortem explained so well and very useful. This is helpful for investors both retail and institutional. Thank you Professor for distilling out the sound wisdom and knowledge from the noise and distractions.
Thanks professor! from Sydney Australia!
Have to say the separation in the run of the mill and the high end usage of AI was insightful as well as the "cute" ones. I am interested in the innovative use cases of AI that can make a small but positive change in people's lives and make them think more. Maybe a red-pill/blue-pill AI is the need of the hour.
Love the analysis prof!
The chip equipment companies such as $ASML falls in the architecture segment right?
Did you watch the video? Ineternet has allowed us to look up anything and decreased our reasoning skills. Try to work it out yourself with the information you already have
Hey, the Valuation playlist's video on P/E multiple has no audio. Could you please re-upload it A better version would be much appreciated. Thanks!
Without doubt, the best video I’ve watched in months. Thank you for such a structured and thoughtful overview.
17:07 deep what??😂
Jk,love you Aswath,thank you for all these wisdom gems❤
Deepseek ended AI monopoly / oligopoly power as it's opensource. I would add to your model assumptions the effect of a US govt clampdown on Nvidia's "Singapore" sales
Great point about the blind spots. Having spent time in silicon valley, it's clear their version of AI abundance is somewhat ignorant of how the economy works
Thanks for an excellent discussion Prof
Deepseek reverted closedAI back to OpenAI. Now anyone can wait for them to spend 100s millions of dollars to train frontier models so we can siphon the model with destillation
The best video and I bow down to your level of clarity and thinking.
Glad that we have a master like you. Wish I had it in my school days.
Insightful and thought provoking as usual. Thanks
Looks like there’s a cut paste error in the mag 6 chart?
Thank you for sharing
Definitely useful......and enjoyable. thanks
You forgot about the chip manufacturers and the manufacturers of the equipment to manufacture chips (your AI players story). Too much hate on physically capital intensive companies!
Waiting for it to
Deepseek does thing at free of cost for which openai asks 200 dollars.
Seeing for first time .
Very interesting how to value a company
25:45 :
The value of the “rest” (which includes gaming) is up about 300 B vs the September valuation , partly compensating the ~40% (or 550 b) drop in the AI business.
Would anyone know why that is the case ?
It's rumored that huawei is developing a chip equal to nvidia's flagship h100 at a significantly lower cost. This could democratize the high end AI market as well.
They are still a few generations behind and if they don't have access to the processes of ASML because of export restrictions it may take a while. On the other hand with tariffs on EU companies by USA and increased isolationism, Europeans may be motivated to lower restrictions to China and have a more balanced approach.
love the remarks you left at the end of the video. thanks aswadh!
Hey, the valuation playlists vidio on P/E multiple has no audio could you re-upload again.
😢
Most reasonable take on this hype nonsensical valuations
the news of curb on AI chips from china could be the another/additional reason too for drop in NVIDIA share price?!
Hi, Didn't Elon say Tesla will need to spend 500 billion dollars on compute for training in the latest earnings call? Then there's other companies who will enter the robotics and humanoid space, this not rule based, I think long term the Nvida story is still justified.
AI for big 7 is already making rewards, I know Big7 companies have been able to cut head count over 1000s because AI does what those head count were doing.
Very interesting thoughts
❤
Thank you professor, for the lesson in common sense and yes I found this video useful.
♥
if this is a shift, then nvida will have a larger market. the current players will still suff their datacenters with chips because they still need to train the models. but additional players will show up because they use better models now on normal consumer cards. maybe this is also a tailwind to amd because there are also competitive chips aiming more on the consumer market
Gosu!
Imo Nvidia is not even worth 1T.
You must be crazy to still hold it at this valuation.
Deepseek required the OpenAi model to be built (see OpenAi lawsuit). The massive spending of OpenAI and others will be required for further development
The distillation of the OpenAi model by Deepseek was allowed if the resulting model was not commercial - this is why the Deepseek was open sourced.
So what, most importantly Deep Seek is free & open source!. It smashed the scam of the big timers & striped them almost naked ! LoL
@@raul298 interesting - so how is there a lawsuit? If they used distillation, do you think they illegally sourced high end Nvidia chips?
1st time comment 😅